IJSPM010209-Kleijnen.

qxd

3/4/2005

6:04 PM

Page 82

82

International Journal of Simulation & Process Modelling, Volume 1, Nos. 1/2, 2005

Supply chain simulation tools and techniques: a survey

Jack P.C. Kleijnen
Department of Information Systems and Management, Center for Economic Research (CentER), Tilburg University, PO Box 90153, 5000 LE Tilburg, The Netherlands E-mail: Kleijnen@UvT.NL
Abstract: The main contribution of this paper is twofold: it surveys different types of simulation for supply chain management it discusses several methodological issues. These different types of simulation are spreadsheet simulation, system dynamics, discrete-event simulation and business games. Which simulation type should be applied, depends on the type of managerial question to be answered by the model. The methodological issues concern validation and verification, sensitivity, optimisation, and robustness analyses. This sensitivity analysis yields a shortlist of the truly important factors in large simulation models with (say) a hundred factors. The robustness analysis optimises the important factors controllable by management, while accounting for the noise created by the important non-controllable, environmental factors. The various methodological issues are illustrated by a case study involving the simulation of a supply chain in the mobile communications industry in Sweden. In general, simulation is important because it may support the quantification of the benefits resulting from supply chain management. Keywords: logistics; performance measurement; Taguchi, risk analysis; uncertainty analysis; screening, sequential bifurcation. Reference to this paper should be made as follows: Kleijnen, J.P.C. (2005) Supply chain simulation tools and techniques: a survey, International Journal of Simulation & Process Modelling, Vol. 1, Nos. 1/2, pp.8289. Biographical note: Jack P.C. Kleijnen is a Professor of Simulation and Information Systems in the Department of Information Management; he is also a member of the Operations Research group of CentER (Center for Economic Research) of Tilburg University. His research concerns computer simulation (especially the statistical design and analysis of simulation experiments), information systems and supply chains. He has been a consultant for several organisations in the USA and Europe, and served on many international editorial boards and scientific committees. He has spent several years in the USA, at universities and private companies. He received a number of international awards. More information can be found on his web page (http://center.kub.nl/staff/kleijnen/), including details on his 200 publications, seminars, simulation course, academic honours, previous employment, education, who is who listings, etc. 1 INTRODUCTION

Simulation analysts may want to quantify the benefits resulting from supply chain management (SCM), in order to support decision making at two levels: the strategic level, including (re)designing a supply chain the operational level, including setting the values of control policies. Kleijnen and Smits (2003) distinguish four simulation types for SCM:

spreadsheet simulation system dynamics (SD) discrete-event dynamic systems (DEDS) simulation business games. Spreadsheets may be part of production control software. SD simulation may explain the bullwhip effect. DEDS simulation may predict fill rate values. Business games may educate and train users (games must be distinguished from gaming; the latter derives analytical solutions including Nash equilibria; see Lau and Lau (2004) for an SCM

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 83

SUPPLY CHAIN SIMULATION TOOLS AND TECHNIQUES: A SURVEY

83

application). In Section 2, I summarise these four simulation types and their role in SCM. From the viewpoint of methodology, I distinguish four types of issues in simulation (in SCM and in other application domains): validation and verification (V & V) sensitivity or what-if analysis optimisation robustness, risk or uncertainty analysis. To address these four methodological issues, a variety of techniques may be used. I, however, focus on the use of statistical methods for the design of experiments (DOE). In Section 3, I describe these methods and illustrate their application through a case study detailed in Kleijnen, Bettonvil and Persson (2004a, 2004b). DOE is important in simulation, because, by definition, simulation is an experimental method; i.e., the analysts experiment with different input values and different model structures (representing different policies, etc.) of the simulation model, treated as a black box (by definition, a black box means that only the inputs and outputs of that box are observed, not its internal operation). A point to note is that some methods do not treat the simulation model as a black box; examples are perturbation analysis and score function methods; see Spall (2003). Unfortunately, these methods require that more mathematical conditions are satisfied, and that analysts are mathematically sophisticated. In practice, simulation is a method that is relatively often used as compared with other quantitative models. Several reasons may explain this popularity: no mathematical sophistication is needed (see the preceding paragraph), multiple responses are natural in simulation (in SCM, these responses may be the fill rate or service percentage, stocks including work in progress or WIP, sales, etc.). These various responses are discussed by Gunasekaran, Patel and McGaughey (2004) and Kleijnen and Smits (2003). Simulation may give insight into the causes and effects of the supply chain performance: which inputs (or factors) significantly affect which outputs? Indeed, simulation can help in understanding causality, as simulation is a methodology that does not treat a system (for example, a supply chain) as a black box (DOE treats the simulation as a black box, as we saw above). For example, modern simulation software may model individual events such as order arrivals and machine breakdowns in great detail; see Kelton, Sadowski and Sturrocks (2004) manual for simulation in the Arena software. Vamanan et al. (2004) compare Arena and other commercial off the shelf (COTS) software; Biswas and Narahari (2004) present object-oriented software for simulation models and other model types of supply chains.) The main contribution of this paper is twofold: it surveys major tools and techniques for the simulation of supply chains it discusses several methodological issues, including a novel methodology for the robust design of supply chains.

The remainder of this paper is organised as follows. The four simulation types that Kleijnen and Smits (2003) distinguish, are discussed in the four separate subsections of Section 2. Section 3 summarises a case study that is detailed in Kleijnen, Bettonvil and Persson (2004a, 2004b); this study illustrates sensitivity analysis, used to derive a shortlist with the most important factors, and robustness analysis, inspired by Taguchi, and bootstrapping to derive a confidence region for the best input values. Section 4 gives the conclusions. A list with 44 references for further study is included.
2 FOUR SIMULATION TYPES FOR SUPPLY CHAIN MANAGEMENT

By definition, a simulation model has the following three characteristics: It is a quantitative, mathematical, computer model. It is a dynamic model; i.e., it has at least one equation with at least one variable that refers to at least two different points in time (examples are difference equations; more examples will follow below). This model is not solved by mathematical analysis; instead, the time paths of the dependent variables (outputs) are computed, given the initial state of the simulated system, and given the values of the exogenous (input) variables. The last aspect implies that simulation does not give a closed form solution. Instead, the simulation analysts experiment with different input values and model structures, to see what happens to the outputso-called sensitivity analysis. Next the analysts may perform V & V, optimisation, and robustness analyses (see Sections 1 and 3). In the following four subsections, I summarise the four simulation types.
2.1 Spreadsheet simulation

Corporate modelling has become popular with the introduction of spreadsheet software; (Plane, 1997; Powell, 1997). Indeed, this type of simulation has made simulation credible for managers. A simple example of an equation that is easy to program through a spreadsheet is:
new inventory = old inventory + production sales. (1)

Equation (1) may be called a bookkeeping equation, a balance equation, a difference equation, etc. Such equations are also part of the more sophisticated simulation types discussed below. Spreadsheets have been used to implement manufacturing resource planning (MRP), which is an important subsystem of SCM; (Sounderpandian, 1989). A recent spreadsheet model of vendor managed inventory (VMI) in supply chains is presented in Disney and Towill (2003). However, this type of simulation is often too simple and unrealistic; DEDS simulation provides a more realistic model (see Section 2.3 below).

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 84

84

J.P.C. KLEIJNEN
Waiting time of job i = max (0, Waiting ti me of job i 1 + Serv ice time of job i 1 Int erarrival time of job i )

2.2 System dynamics (SD)

Forrester (1961) developed industrial dynamics, which he later extended and called system dynamics. In fact, Forrester has already developed a model for the following supply chain, without using the term supply chain. His supply chain (which is theoretical, academic) has four links, namely retailer, wholesaler, distributor and factory. He examines how these links react to deviations between actual and target inventories. He finds that common sense strategies may amplify fluctuations in the demand by final customers, up in the supply chain. Much later, Lee, Padmanabham and Whang (1997) identified this amplification as one of the bullwhip effects; see also Disney and Towill (2003). A recent case study is provided by Higuchi and Troutt (2004), who use SD to model the supply chain for the Japanese pet-toy called Tamagotchi. Spengler and Schroter (2003) also uses SD and study spare parts in closed-loop supply chains at Agfa-Gevaerts. Ashayeri and Keij (1998) model the distribution chain of Edisco (the European distribution arm of the US company Abbott Laboratories). Reviews of SD simulation of SCM are Angerhofer and Angelidis (2000), Beamon (1998), Otto and Kotzab (2003), and Van der Pol and Akkermans (2000). From a methodological viewpoint, SD views companies as systems with six types of flows, namely materials, goods, personnel, money, orders and information (examples of these input flows are production and sales; example output flows are fill rate and average WIP). Besides flows, SD distinguishes stocks (for example, WIP at a given point in time). SD assumes that managerial control is realised through the changing of rate variables (for example, production and sales rates), which change flows, and hence stocks. A crucial role in the SD worldview is played by the feedback principle; i.e., a manager compares a target value for a specific performance metric with its realisation, and in case of undesirable deviationthe manager takes corrective action. An example equation is
Inventory.K = Inventory.J DT Sales_rate.JK) (Production_rate.JK (2)

(3)

DEDS simulation has the following two characteristics: It represents individual events (for example, the arrival of an individual customer order; see Equation (3)), whereas SD has a much more aggregated view including flows. It incorporates uncertainties (for example, customer orders arrive at random points in time; see again Equation (3); machines break down at random points of time, and require random repair times). The other three types of simulation models remain relevant, even when eliminating randomness! For example, most SD models have no randomness, and yet their behaviour remains counter-intuitive because of the nonlinear feedback loops. (Most econometric models are also sets of non-stochastic, nonlinear difference equations.) Also see Gaonkar (1977). For more details on DEDS simulation I refer to the many textbooks on this type of simulation, including the most popular (83,000 copies sold) one by Law and Kelton (2000). DEDS simulation is an important method in SCM. For example, Banks et al. (2002) survey many SCM simulation studies, at IBM and Virtual Logistics, and they discuss strategic and operational SCM, distributed SCM simulation, commercial packages for SCM simulation, etc. Indeed, DEDS simulation is already part of the MRP/ERP toolbox for quantifying the costs and benefits of strategic and operational policies (ERP: Enterprise Resource Planning); see Vollmann, Berry and Whybark (1997). In Section 3, I shall discuss a recent example of DEDS simulation that models three alternative designs for a supply chain in the mobile communications industry in Sweden that is centred on the Ericsson company.
2.4 Business games

where Sales_rate.JK denotes the sales rate during the interval between the points of time J and K; DT denotes the length of that interval; etc. For more details on SD, I refer to a recent SD textbook such as Sterman (2000), which has 982 pages!
2.3 Discrete-event dynamic system (DEDS) simulation

A DEDS simulation is more detailed than the preceding two simulation types, as is illustrated by comparing Equations (1) and (2) with the following example DEDS equation:

It is relatively easy to simulate technological and economic processes, but it is much more difficult to model human behaviour. A solution is to let managers themselves operate within the simulated world, which may consist of a supply chain and its environment. Such an interactive simulation is called a business or management game. Games may be used for both educational and research goals. For their education usage, I refer to Riis, Smeds and Van Landeghem (2000) and Ten Wolde (2000). For research usage, I refer to Kleijnen (1980). For example, Kleijnen (1980, pp.157186) uses an IBM management game to quantify the effects of information accuracy on return on investment (ROI). Another example is the use of games to study the confidence that managers have in their decisions. More recent references are given by Kleijnen and Smits (2003) and Riis, Smeds and Landeghem (2000).

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 85

SUPPLY CHAIN SIMULATION TOOLS AND TECHNIQUES: A SURVEY

85

There are strategic and operational games: Strategic games include several teams of players who represent companies that compete with each other in the simulated world. These players interact with the simulation model during (say) five to ten rounds. The simulation model may be an SD model; a famous example is the beer game, which illustrates the bullwhip effect; see Simchi-Levi, Kaminsky and Simchi-Levi (2003); Sodhi (2001), and Sterman (2000). The game may also be a corporate, economic, business model that illustrates the effects of prices, sales promotion and R&D decisions on profits; see Kleijnen (1980, pp.157186). Also see Riis, Smeds and Landeghem (2000). Operational games include a single team, which may consist of one or more players, interacting with the simulation model either during several rounds or in real time. This are games against nature. Examples are games for training in production scheduling; see again Riis, Smeds and Landeghem (2000).
2.5 The roles of different simulation types in SCM

optimisation and robustness analysis (which were mentioned in Section 1). The study concerns the strategic level of SCM; it uses DEDS simulation.
3.1 Overview

Which of the four simulation types is applied in SCM depends on the problem to be solved. For example, SD aims at qualitative insight (not exact forecasts); for example, SD can demonstrate the bullwhip effect. DEDS simulation can quantify fill rates, which are random variables. Games can educate and train users, since the players are active participants in the simulated world. Moreover, games can be used in research to study the effects of qualitative factors (such as type of decision support system, DSS) on profits, etc. For brevitys sake, I refer back to the publications that were discussed in the four preceding subsections.
3 V & V, SENSITIVITY, OPTIMISATION AND ROBUSTNESS ANALYSES: A CASE STUDY

In this section, I summarise a case study that is detailed in Kleijnen, Bettonvil and Persson (2004a, 2004b). This study illustrates the importance of V & V, sensitivity analysis,

The case study consists of three simulation models that represent three alternative designs for a supply chain. Figure 1 illustrates that a newer configuration has fewer operations and tests. Figure 2 shows one of the simulation modelsbuffers (inventories) are located before and after every test station and operation; products are transported between all operations and test stations. The output is the steady-state mean costs of the total supply chain. Details are given by Persson and Olhager (2002). Kleijnen, Bettonvil and Persson (2004a) derive a shortlist with the most important factors; this process is also called screening. They apply a method called sequential bifurcation (SB). I shall summarise this study in Section 3.3. Next, Kleijnen, Bettonvil and Persson (2004b) derive a robust solution; i.e. they find appropriate values for the factors that management can control, while accounting for the randomness of the environmental factors. Their solution is inspired by Taguchis approach for designing robust physical products. This means that the important factors are divided into controllable and environmental factors. Kleijnen, Bettonvil and Persson (2004b) systematically investigate these controllable factors (using a reduced so-called central composite design). They randomly combine the environmental factors into scenarios using latin hypercube sampling, LHS. Then they estimate the controllable factor values that minimise the outputs expected value and variance, respectively. A confidence region for these optima is derived through bootstrapping. This confidence region can be used to select a (compromise) robust solution. I shall summarise that study in Section 3.4. Note: The SCM literature distinguishes between robustness and flexibility. A flexible supply chain can react to a changing environment by adapting its operations. A robust supply chain keeps its design fixed, and can still accommodate many changes in its environment. So the two

Figure 1 The three supply chain structures: (a) the old, (b) the current and (c) the next generation

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 86

86

J.P.C. KLEIJNEN

Figure 2 The simulation model of the current supply chain structure

concepts focus on operational and strategic decisions respectively. Also see Van Landeghem and Vanmaele (2002) and Zhang, Vonderembse and Lim (2003). The effects of flexibility on supply chain performance are evaluated through simulation by Garavelli (2003). Note: Robustness is also important outside DEDS simulation. For example, deterministic simulationvia non-linear difference equationsfor computer-aided engineering (CAE) often aims at the optimal design of products such as automobiles, airplanes, etc.; see Simpson et al. (2001). The coefficients of these equations are not exactly known, and the resulting solution cannot be exactly implemented, so the optimal solution should be robust. In mathematical programming, the search for robust solutions has made most progress and reached maturity, I think; see Ben-Tal and Nemirovski (2000).
3.2 Validation and verification (V & V)

The simulation model for the old supply chain depicted in Figures 1 and 2 was validated and verified through discussions with Ericsson engineers; see again Persson and Olhager (2002). This establishes face validity; see Law and Kelton (2000). Quantitative V & V may use DOE to check whether the estimated effects of changing inputs of the simulation model agree with the experts qualitative knowledge about the system. A simple example is a queuing simulation: does an increase in simulated traffic give an increased average waiting time? A case study is reported by Kleijnen (1995): does the DEDS simulation give correct signs for the estimated effects of inputs (such as sonar tilt angle) on the output (detection of mines at the sea bottom)? If data on the real-world output are available, then real and simulated outputs may be compared statistically. An overview of the role of different statistical techniquesdependent on the availability of datain V & V is given in Kleijnen (1999).
3.3 Screening through sequential bifurcation

The SB method simulates relatively few scenarios (factor combinations); for example, Kleijnen, Bettonvil and Persson (2004a) simulate only 42 scenarios to find the 11 most important factors among the 92 potentially important factors. To realise this efficiency, SB uses two basic assumptions: A first-order polynomialpossibly augmented with two-factor interactionscan adequately approximate the input/output (I/O) behaviour of the underlying simulation model. (Such approximations are also called metamodels, because they model the underlying simulation models I/O behaviour.) The signs (or directions) of all main effects are known, so factors can be defined such that all main effects are non-negative (otherwise, main effects might compensate each other). Because of the first assumption, SB simulates only two values per factor, namely a high and a low value. (In the case study, most factors change by 5% of the base value; a few other factors by 25%.) Estimation of main effects unbiased by two-factor interactions is enabled by the so-called foldover design, which doubles the number of scenarios that would be simulated in case the polynomial were known to have first-order effects only. To estimate the statistical significance of the estimated effects, each scenario needs replication by using different, non-overlapping pseudo-random numbers (PRN). In the case study, the number of replicates is selected to be five. In the case study, SB turns out to give only 11 important factors for the old model, nine for the current model and seven for the next generation model. In all three simulation models, factor number 92 is the most important factor; this is the demand for product 1, which accounts for 90% of total demand. The other important factors represent yield and transportation. See Kleijnen, Bettonvil and Persson (2004a) for details.
3.4 Robustness analysis

The total number of potentially important factors in the three simulation models is 92 in the old model, 78 in the current model and 49 in the next generation model. The most important factor is defined as the one with the highest main effectalso called the first-order effect; see the SB assumptions below.

Taguchi is the Japanese engineer who designed cars (for Toyota) that operate satisfactorily in many environments; see Taguchi (1987). His method is applied to simulation by Al-Aomar (2002) and Tsai (2002). Next, Kleijnen, Bettonvil and Persson (2004b) use Taguchis view

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 87

SUPPLY CHAIN SIMULATION TOOLS AND TECHNIQUES: A SURVEY

87

but not his statistical methods, because simulation experiments enable the exploration of many more factors and scenarios than are possible in real-life experiments. Kleijnen, Bettonvil and Persson (2004b) try to minimise expected cost (as in classic optimisation), but also consider cost variance due to environmental disturbances (as Taguchi proposes). For illustration, I focus on the current model. After the SB screening (Section 3.3), only three important controllable factors remain (and six important environmental factors). The challenge is to optimise these controllable factors, denoted by (say) xj ( j = 1, , k ). A second-order polynomial approximation of the I/O behaviour of the simulation model is:
yi = 0 + j = 1 j xi ; j + j = 1 j ; j xi2 j ; + j = 1 j '= j + 1 j ; j ' xi ; j xi ; j ' + ei (i = 1, 2, ....)
k 1 k k k

conflicting optimal values when considering both outputs. Yet, these optimal input values are only estimates, so maybe the truly optimal inputs are the same for both outputs (mean and variance)? To answer this question, a confidence region is needed for the optimal values. Standard confidence regions do not hold, because the estimated optimal simulation inputs are nonlinear functions of the simulation outputs and the corresponding regression estimates. Yet, a confidence region can be computed through bootstrapping. Assuming that the estimated regression parameters corresponding with Equation (4) are normally distributed, the bootstrap resamples from a multivariate normal distribution with a vector of means equal to the original estimates, and with a covariance matrix equal to the estimated matrix that is a standard output of regression software. The resulting * bootstrapped estimated regression parameters (say) give
estimated optimal inputs x* . Bootstrapping repeats this sampling (say) 1,000 times, to get a confidence region for the values that minimise the mean and the variance of the output respectively. Using this region, management may select a robust solution that satisfies their preferences. Note: Bootstrapping is computationally inexpensive, compared with the computer effort required to generate the simulation output. Finally, comparing the optimum solution for the controllable factors accounting for many environmental scenarios that are generated through LHS, and the solution accounting for a single scenario, namely the base scenario, suggests that risk considerations do make a difference; see Kleijnen, Bettonvil and Persson (2004b).
4 CONCLUSIONS

(4)

with the overall mean 0, the first-order effects j, the interactions (cross-products) j; j and the error termeiwhich represents noise caused by both the PRN and the lack of fit of the approximation in Equation (4) in scenario i. To estimate the coefficients of the second-order polynomial in Equation (4), Kleijnen, Bettonvil and Persson (2004b) augment the 2k full factorial design with a one-factor-at-a-time design, so that each factor is observed at more than two values. For the environmental factors, robustness analysis is not interested in a functional relationship like Equation (4). Following Taguchi, Kleijnen, Bettonvil and Persson (2004b) treat these factors as noise. Unlike Taguchi, they sample environmental scenarios through LHS; McKay, Beckman and Conover (1979). Further, as in a Taguchian design, the design (inner array) for the controllable factors is crossed (combined) with the design (outer array) for the environmental factors. This crossed design is not replicated (unlike SB; see Section 3.3): the standard error in each design point due to pure replication (using different PRN) turns out to be much smaller than the standard error across the environmental scenarios. The optimisation of Equation (4) for different environmental scenarios should account for the box constraints on the inputs: only changes of 5% and 25% are allowed. Mathematically, these constraints are incorporated through Lagrangian multipliers, which quantify the shadow prices of the constraints. All controllable factors turn out to minimise the mean costs when they are set at their lower boundary values. (Because these bounds are specified rather arbitrarily, I raise the question whether these boundary values should not be revised, and followed-up, by a new analysis?) Next, the response variable y in Equation (4) is replaced by the output variance. One factor again has its optimal value at its lower boundary. The other factors, however, have

In this paper, I surveyed four types of simulation, and discussed four methodological issues. These four simulation types are spreadsheets, SD, DEDS and gaming. I explained how different types can answer different questions in SCM. For example, SDpossibly run as a gamesuffices for demonstrating the bullwhip effect to supply chain stakeholders. DEDS is needed to estimate the probability of realising a required fill rateespecially in a turbulent environment. Once a simulation model has been built, it is necessary to perform V & V, sensitivity analysis, optimisation and robustness analysis of that model. V & V may use statistical techniques, including DOE. Sensitivity analysis serves several goals: it provides insight into the behaviour of the supply chain (including interactions between factors), and gives a shortlist of critical factors. Optimising the critical control factors may support business process redesign (BPR). In practice, it is more important to find robust solutions than the optimal solution. This paper summarised a novel methodology for searching for such a robust solution. This solution gives values for those factors that management can control, while accounting for the randomness of the environmental factors.

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 88

88

J.P.C. KLEIJNEN Higuchi, T. and Troutt, M.D. (2004) Dynamic simulation of the supply chain for a short life cycle product-lessons from the Tamagotchi case, Computers & Operations Research, Vol. 31, pp.10971114. Kelton, W.D., Sadowski, R.P. and Sturrock, D.T. (2004) Simulation with Arena, Third Edition, McGraw-Hill, Boston, MA. Kleijnen, J.P.C. (1980) Computers and Profits: Quantifying Financial Benefits of Information, Addison-Wesley, Reading, MA. Kleijnen, J.P.C. (1995) Case study: statistical validation of simulation models, European Journal of Operational Research, Vol. 87, No. 1, pp.2134. Kleijnen, J.P.C. (1999) Validation of models: statistical techniques and data availability, in Farrington, P.A., Nembhard, H.B., Sturrock, D.T. and Evans, G.W. (Eds.): Proceedings of the 1999 Winter Simulation Conference, pp. 647654. Kleijnen, J.P.C. and Smits, M.T. (2003) Performance metrics in supply chain management, Journal of the Operational Research Society, Vol. 54, No. 5, pp.507514. Kleijnen, J.P.C., Bettonvil, B. and Persson, F. (2004a) Finding the important factors in large discrete-event simulation: sequential bifurcation and its applications, in Dean, A.M. and Lewis, S.M. (Eds.): Screening, Springer-Verlag, New York. Kleijnen, J.P.C., Bettonvil, B. and Persson, F. (2004b) Robust solutions for supply chain management: Simulation and risk analysis of the Ericsson case study, Working Paper (preprint: http://center.kub.nl/staff/kleijnen/papers.html). Lau, A.H.L. and Lau, H. (2004) Some two-echelon supply chain games: improving from deterministic-symmetric-information to stochastic-asymmetric-information models, European Journal of Operational Research (forthcoming). Law, A.M. and Kelton, W.D. (2000) Simulation Modeling and Analysis, Third Edition, McGraw-Hill, Boston, MA. Lee, H.L., Padmanabhan, V. and Whang, S. (1997) The bullwhip effect in supply chains, Sloan Management Review, Spring, Vol. 38, pp.93102. McKay, M.D., Beckman, R.J. and Conover, W.J. (1979) A comparison of three methods for selecting values of input variables in the analysis of output from a computer code, Technometrics, Vol. 21, pp.239245. Otto, A. and Kotzab, H. (2003) Does supply chain management really pay? Six perspectives to measure the performance of managing a supply chain, European Journal of Operational Research, Vol. 144, pp.306320. Persson, F. and Olhager, J. (2002) Performance simulation of supply chain designs, International Journal of Production Economics, Vol. 77, pp.231245. Plane, D.R. (1997) How to build spreadsheet models for production and operations management, OR/MS Today, Vol. 24. No. 2, pp.5054. Powell, S.G. (1997) Leading the spreadsheet revolution, OR/MS Today, Vol. 24, pp.810. Riis, J.O, Smeds, R. and Van Landeghem, R. (Eds.) (2000) Games in Operations Management, Kluwer, Boston, MA. Simchi-Levi, D., Kaminsky, P. and Simchi-Levi, E. (2003) Designing and Managing the Supply Chain: Concepts, Strategies, and Case Studies, Second Edition, Irwin/McGrawHill, Boston, MA.

This robustness methodology was inspired by Taguchis approach. Technically, however, other designs were proposed, for example, LHS. Moreover, a confidence region for the estimated optima was proposed, based on bootstrapping; management may use that region to select a compromise, robust solution. The methodological issues were illustrated through a case study, namely simulation models of different supply chain configurations for Ericsson in Sweden.

REFERENCES

Al-Aomar, R. (2002) A robust simulation-based multicriteria optimisation methodology, in Ycesan, E., Chen, C.H., Snowdon, J.L. and Charnes, J.M. (Eds.): Proceedings of the 2002 Winter Simulation Conference, pp.19311939. Angerhofer, B.J. and Angelidis, M.C. (2000) System dynamics modelling in supply chain management: research view, in Joines, J.A. et al. (Eds.): Proceedings of the 2000 Winter Simulation Conference, pp.342351. Ashayeri, J. and Keij, R. (1998) Global business process re-engineering: a system dynamics based approach, International Journal of Operations and Production Management, Vol. 18, pp.817831. Banks, J., Buckley, S., Jain, S. and Lendermann, P. (2002) Panel session: opportunities for simulation in supply chain management, in Ycesan, E., Chen, C.H., Snowdon, J.L. and Charnes, J.M. (Eds.): Proceedings of the 2002 Winter Simulation Conference, pp. 165211658. Beamon, B.M. (1998) Supply chain design and analysis: models and methods, International Journal of Production Economics, Vol. 55, pp.281294. Ben-Tal, A. and Nemirovski, A. (2000) Robust solutions of linear programming problems contaminated with uncertain data, Mathematical Programming, Vol. 88, pp.411424. Biswas, S. and Narahari, Y. (2004) Object oriented modelling and decision support for supply chains, European Journal of Operational Research, Vol. 153, pp.704726. Disney, S.M. and Towill, D.R. (2003) MRP on spreadsheets: a do-it-yourself alternative for small firms, Production and Inventory Management Journal, Second Quarter, pp.611. Forrester, J.W. (1961) Industrial Dynamics, MIT Press, Cambridge, MA. Gaonkar, R. (1977) Comparative analysis of systems dynamics modeling and econometric modelling, in Hamza, M.H. (Ed.): Simulation 1977 Proceedings of the International Symposium, Act Press, Anaheim, CA. Garavelli, A.C. (2003) Flexibility configurations for the supply chain management, International Journal of Production Economics, Vol. 85, pp.141153. Gunasekaran, A., Patel, C. and McGaughey, R.E. (2004) A framework for supply chain performance measurements, International Journal of Production Economics, Vol. 87, No. 3, pp.333347.

IJSPM010209-Kleijnen.qxd

3/4/2005

6:04 PM

Page 89

SUPPLY CHAIN SIMULATION TOOLS AND TECHNIQUES: A SURVEY Simpson, T.W., Peplinski, J., Koch, P.N. and Allen, J.K. (2001) Metamodels for computer-based engineering design: survey and recommendation, Engineering with Computers, Vol. 17, No. 2, pp.129150. Sodhi, M.S. (2001) Applications and opportunities for operations research in internet-enabled supply chains and electronic marketplaces, Interfaces, Vol. 31, pp.5669. Sounderpandian, J.D.B.A (1989) MRP on spreadsheets: a do-it-yourself alternative for small firms, Production and Inventory Management Journal, Second Quarter, pp.611. Spall, J.C. (2003) Introduction to Stochastic Search and Optimisation; Estimation, Simulation and Control, Wiley, Hoboken, NJ. Spengler, T. and Schrter, M. (2003) Stategic management of spare parts in closed-loop supply chainsa system dynamics approach, Interfaces, Vol. 33, No. 6, pp.717. Sterman, J.D. (2000) Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw-Hill, Boston, MA. Taguchi, G. (1987) System of Experimental Designs, UNIPUB/Krauss International, White Plains, NY, Vol. 1/2. Ten Wolde, H. (2000) Building blocks of education, OR/MS Today, Vol. 27, pp.12. Tsai, C.S. (2002) Evaluation and optimisation of integrated manufacturing system operations using Taguchis experiment design in computer simulation, Computers & Industrial Engineering, Vol. 43, No. 3, pp.591604.

89

Vamanan, M., Wang, Q., Batta, R. and Szczerba, R.J. (2004) Integration of COTS software products ARENA & CPLEX for an inventory/logistic problem, Computers & Operations Research, Vol. 31, pp.533547. Van der Pol, J.M and Akkermans, H.A. (2000) No one in the drivers seat: an agent-based modelling approach to decentralised behaviour in supply-chain co-ordination, Eleventh International Working Seminar on Production Economics, Pre-prints, (Igls, Austria), Vol. 3, p.621642. Van Landeghem, H. and Vanmaele, H. (2002) Robust planning: a new paradigm for demand chain planning, Journal of Operations Management, Vol. 20, pp.769783. Vollmann, T.E., Berry, W.L. and Whybark, D.C. (1997) Manufacturing Planning and Control Systems, Fourth Edition, Irwin/McGraw-Hill, New York. Zhang, Q., Vonderembse, M.A. and Lim, J.S. (2003) Manufacturing flexibility: defining and analysing relationships among competence, capability, and customer satisfaction, Journal of Operations Management, Vol. 21, pp.173191.